Details
-
Improvement
-
Status: Open
-
Minor
-
Resolution: Unresolved
-
None
-
Docs Required, Release Notes Required
Description
HashUtils.Hash64Internal can benefit from SIMD acceleration on large values.
Currently it splits the data into blocks of 16 bytes (2 longs), but we can process larger blocks with Vector<long> (4 pairs at once with AVX2, 8 pairs at once with AVX512).
Example: https://gist.github.com/dnbaker/0fc1d4edbbdb24069eb063dc2559e4f5